Frequency Clustering Analysis for Resting State Functional Magnetic Resonance Imaging Based on Hilbert-Huang Transform

Objective: Exploring resting-state functional networks using functional magnetic resonance imaging (fMRI) is a hot topic in the field of brain functions. Previous studies suggested that the frequency dependence between blood oxygen level dependent (BOLD) signals may convey meaningful information regarding interactions between brain regions. Methods: In this article, we introduced a novel frequency clustering analysis method based on Hilbert-Huang Transform (HHT) and a label-replacement procedure. First, the time series from multiple predefined regions of interest (ROIs) were extracted. Second, each time series was decomposed into several intrinsic mode functions (IMFs) by using HHT. Third, the improved k-means clustering method using a label-replacement method was applied to the data of each subject to classify the ROIs into different classes. Results: Two independent resting-state fMRI dataset of healthy subjects were analyzed to test the efficacy of method. The results show almost identical clusters when applied to different runs of a dataset or to different datasets, indicating a stable performance of our framework. Conclusions and Significance: Our framework provided a novel measure for functional segregation of the brain according to time-frequency characteristics of resting state BOLD activities.